What do you mean that the search engines contain minimal amount of site’s data? Obviously it needs to index all contents to make it searchable. If you search for keywords within an article, you can find the article, therefore all of it needs to be indexed.

Indexing is nothing more than “presenting data to the algorithm” so it’d be against the law to index a site under your proposed legislation.

Wrong. The infringement is in obtaining the data and presenting it to the AI model during the training process. It makes no difference that the original work is not retained in the model’s weights afterwards.

This is an interesting take, I’d be inclined to agree, but you’re still facing the problem of how to distinguish training AI from indexing for search purposes. I’m afraid you can’t have it both ways.